Skip to content

Prompt Engineering Foundations · Your prompt pattern library · lesson 13 of 16 · 12 min

Patterns for analysis and brainstorming (19-37)

Analysis patterns

Analysis is where AI can save hours, as long as you give it the material and check its conclusions.

19. The summary for a reader

Summarise this for [reader] who has [2 minutes / no background].
Focus on [what matters to them]: "[text]"

20. The key points and gaps

List the key points in this document, then list any important
questions it doesn't answer: "[text]"

21. The compare

Compare [option A] and [option B] for [my situation] in a table:
cost, effort, risks, benefits, best for.

22. The pros and cons with weight

List the pros and cons of [decision] for [who]. Mark each as
high, medium or low importance and explain briefly.

23. The theme finder

Here are [customer comments / survey answers]. Group them into themes,
count how many comments fit each theme, and give one example quote each.

24. The explain the numbers

Here is a small table of [data]. Describe the main patterns in plain
language, and tell me what you can't conclude from this data alone.

25. The devil's advocate

Here is my plan: [plan]. Argue against it as strongly as you can.
What would a sceptic say?

26. The assumption check

What assumptions am I making in this plan? Which ones are riskiest
if they're wrong? [plan]

27. The extract

From the text below, extract every [date / name / price / action]
into a list. If something is unclear, mark it "unclear": "[text]"

28. The decision helper

I'm deciding between [options]. My priorities, in order, are [1, 2, 3].
Ask me any questions you need, then recommend one and explain why.

Brainstorming patterns

Brainstorming is where AI's speed shines. Quantity first, then filter.

29. The big list

Give me 20 ideas for [goal] for [audience]. One line each. Include
some unusual ones.

30. The constraint twist

Give me 10 ideas for [goal] that cost less than [budget] and can be done
in [time].

31. The perspective shift

How would [a teenager / a busy parent / a retiree / a competitor] see
[product or idea]? Give 3 ideas from each perspective.

32. The combine

Combine ideas from [field A] and [field B] to generate 10 new ideas for [goal].

33. The name generator

Suggest 15 names for [product / event / business]. Criteria: [short,
easy to spell, works in English and Arabic]. Explain the idea behind each.

34. The build on mine

Here are my 3 ideas: [ideas]. Build on each with 3 variations, then
suggest 3 new ones in a different direction.

35. The filter and rank

From the ideas above, pick the best 5 for [criteria] and rank them,
with one sentence each on why.

36. The what could go wrong

For my idea [idea], list the 5 most likely ways it could fail and how
to prevent each.

37. The first step

For the top idea, what is the smallest, cheapest test I could run
this week to see if it works?

Worked example: from comments to action

A café owner in Manchester pastes 40 online reviews and uses patterns 23, 26 and 29:

  1. Theme finder: themes include slow weekend service, loved pastries, limited seating, and friendly staff.
  2. She checks the counts against a few reviews herself (the AI's counts can be off, so spot-check).
  3. Assumption check on her plan to extend opening hours: "You're assuming the complaints about seating are from people who would come at later hours."
  4. Big list: 20 ideas to reduce weekend waits; she picks a pre-order option and a second till.

Important cautions for analysis

  • Give it the data. Without your real material, the AI will produce generic analysis that sounds specific.
  • Check counts and numbers. AI can miscount or miscalculate, especially with long lists. Spot-check, or use a spreadsheet for exact numbers.
  • Correlation is not cause. If the AI says "sales rose because of the campaign", ask what else could explain it.
  • You own the decision. Use AI to widen your thinking, not to replace your judgement.

Hands-on: a mini analysis you can finish in 15 minutes

Paste 15 to 30 real comments (reviews, survey answers or feedback, with names and contact details removed) and run this three-step sequence:

Step 1: Group these comments into 4-6 themes. For each theme give a
count, a one-line description and one short example quote. Put any
comment that fits no theme under "Other".

Step 2: Which theme would most improve customer satisfaction if fixed?
Give your reasoning and say what you can't tell from this data alone.

Step 3: Give me 10 low-cost ideas to fix that theme, then rank the top 3
by effort and likely impact.

Then do the human part: pick three comments at random and check they were placed in a sensible theme, and recount one theme yourself.

When to switch on a thinking mode

For analysis, a thinking or reasoning mode is often worth the extra wait when:

  • the question has several steps ("compare three suppliers across price, delivery and risk, then recommend one");
  • numbers must be combined ("work out the break-even point");
  • you want the assistant to challenge a plan properly (devil's advocate, assumption check).

For quick theme-finding or simple summaries, the normal mode is usually enough. If your assistant can run code or analyse uploaded spreadsheets, ask it to calculate rather than estimate, and check that it did.

Brainstorming with better spread

AI brainstorms can feel samey. Three fixes:

- "Give me ideas in 4 different categories: cheap, bold, digital, community."
- "Include 3 ideas that most businesses in my industry would dismiss."
- "Don't include anything that needs a budget above 500 GBP."

Measuring an analysis prompt

A good analysis prompt produces conclusions you can trace back to the data. If you cannot find the comments behind a theme, or the counts are clearly wrong, the analysis is not usable yet. Ask for quotes and counts every time; they make checking fast.

Video lecture: Patterns for analysis and brainstorming (19-37)

Patterns for analysis and brainstorming (19-37) — Analysis patterns Analysis is where AI can save hours, as long as you give it the material and check its conclusions. 19. The summary for a reader 20…

12 chapters · about 7 minutes · captions and full transcript below.

  1. Patterns for analysis and brainstorming
  2. Why analysis and brainstorming?
  3. Analysis patterns
  4. More analysis patterns
  5. When to think harder
  6. Brainstorming patterns
  7. Better spread
  8. Worked example: café reviews
  9. A 15-minute mini analysis
  10. Example 1: choosing a phone
  11. Example 2: café survey (illustrative)
  12. Recap

Lecture transcript

Patterns for analysis and brainstorming

A café owner pastes forty online reviews into an assistant, and fifteen minutes later she knows exactly what to fix first. In this lecture you will learn the analysis and brainstorming patterns that make that possible, when to switch on a thinking mode for analysis, and the checks that keep AI conclusions honest. By the end, you will turn messy feedback and blank-page moments into clear, ranked actions.

Why analysis and brainstorming?

Why does this matter? Because most businesses sit on feedback they never read properly: reviews, surveys, support emails, sales notes. Not because it is unimportant, but because sorting it takes hours. AI can do the first sort in minutes. And when you are stuck for ideas, it can produce twenty options before you have finished your coffee. The catch is that analysis must stay tied to your real data, and ideas must be filtered by your real constraints. That is what these patterns build in.

Analysis patterns

Analysis patterns first. Summary for a reader: summarise this for my manager who has two minutes, focusing on what matters to them. Key points and gaps: list the key points, then the important questions it does not answer. Compare: put option A and option B in a table by cost, effort, risks and benefits. Theme finder: group these comments into themes, count each, and give one example quote. And the assumption check: what assumptions am I making, and which are riskiest if wrong?

More analysis patterns

Three more analysis patterns are worth knowing. The extract: pull every date, name, price or action into a list, marking anything unclear as unclear. The devil's advocate: here is my plan, argue against it as strongly as you can. And the decision helper: I am deciding between these options, my priorities in order are these, ask me questions, then recommend one and explain why. That last one is a great fit for a thinking mode, because it juggles several priorities at once.

When to think harder

When should you switch on a thinking mode for analysis? When the question has several steps, like comparing three suppliers on price, delivery and risk before recommending one. When numbers must be combined, like working out a break-even point. And when you want a plan challenged properly. For quick theme-finding or simple summaries, the normal mode is usually enough. And if your assistant can run code on an uploaded spreadsheet, ask it to calculate, not estimate, and check that it did.

Brainstorming patterns

Now brainstorming, where AI's speed shines. Quantity first, then filter. The big list: give me twenty ideas, one line each, including some unusual ones. The constraint twist: ten ideas under a set budget that can be done this month. The perspective shift: how would a teenager, a busy parent or a competitor see this? Build on mine: here are my three ideas, give three variations each, then three in a new direction. Then filter and rank: pick the best five for my criteria. And finally, the first step: what is the smallest, cheapest test I could run this week?

Better spread

AI brainstorms can feel samey. Three fixes help. Ask for ideas in named categories, like cheap, bold, digital and community. Ask for three ideas most businesses in your industry would dismiss. And set a hard constraint, such as nothing that needs a budget above five hundred pounds. Constraints and categories force variety.

Worked example: café reviews

Here is the café in Manchester. The owner pastes forty reviews and uses the theme finder. Themes include slow weekend service, loved pastries, limited seating and friendly staff. She spot-checks the counts against a few reviews herself, because AI counts can be off. The assumption check on her plan to extend opening hours points out that she is assuming the seating complaints come from people who would visit later. Then the big list gives twenty ideas to cut weekend waits, and she picks pre-ordering and a second till. Analysis you can trace back to the data.

A 15-minute mini analysis

Let us walk through a fifteen-minute mini analysis. Paste fifteen to thirty real comments with personal details removed. Step one: group them into four to six themes, with counts, a one-line description and an example quote each, and an other bucket. Step two: which theme would most improve satisfaction if fixed, and what cannot be concluded from this data? Step three: ten low-cost ideas for that theme, with the top three ranked by effort and impact. Then do the human part: check three random comments landed in sensible themes, and recount one theme yourself.

Example 1: choosing a phone

A simple example first, with the compare pattern. You are choosing between two phones. Ask: compare these two phones for someone who mainly takes photos of their children and wants the battery to last all day, in a table with price, camera, battery, storage and best for. Then the devil's advocate: argue against my favourite. The table organises the facts, and the counter-argument stops you falling for one shiny feature. Note, though, that specs and prices change, so check them on the manufacturers' pages.

Example 2: café survey (illustrative)

Now a business scenario, with illustrative numbers. Sophie runs a café in Leeds and has two hundred survey responses from a loyalty-card campaign. She uses the theme finder: group these into themes with counts and a quote each. Top themes: slow service at lunchtime, about fifty mentions; love for the pastries, about forty; and requests for oat milk, about thirty. She spot-checks ten responses and finds the counts are roughly right. Then the assumption check on her plan to hire another lunchtime barista: what am I assuming? It points out she has not checked whether the slowness is staff or the till. Then the big list: twenty ways to speed up lunch. She tests a pre-order option first because it is cheapest. Illustrative numbers, real method.

Recap

To recap. Give the AI your real data, ask for quotes and counts so you can check, and keep the decision yours. Correlation is not cause, so if the AI says sales rose because of a campaign, ask what else could explain it. Try this now: paste fifteen to thirty real comments with personal details removed, run the three-step sequence from the lesson, then spot-check three placements and recount one theme yourself. Next lesson: patterns for learning and planning.

Key takeaways

  • Analysis patterns: summary for a reader, key points and gaps, compare, weighted pros and cons, themes, extract, assumption check.
  • Brainstorming patterns: big list, constraint twist, perspective shift, combine, build on mine, filter and rank, first step.
  • Generate many ideas first, then filter with explicit criteria.
  • Give the AI your real data, spot-check counts and numbers, and keep the decision yours.

Try it

Paste 20 real comments (reviews, survey answers or feedback, with personal details removed) and use the theme finder. Spot-check three counts, then use 'first step' on the top improvement idea.